Download the syntax of lexical reciprocal constructions

THE SYNTAX OF LEXICAL
RECIPROCAL CONSTRUCTIONS
Peter Hurst
University of Melbourne
Proceedings of the LFG10 Conference
Miriam Butt and Tracy Holloway King
(Editors)
2010
CSLI Publications
http://csli-publications.stanford.edu/
290
1 Abstract1
Many languages are able to productively form two types of reciprocal
constructions from a transitive base: a monadic construction which groups
the participants in the relation in the subject NP while losing an object NP;
and a dyadic construction which creates a symmetric situation by placing one
participant in a subject NP and another in a comitative phrase (also with the
corresponding loss of an object). I show that the syntax of these constructions
in Swahili (as well as in many other languages) can be understood as a
natural consequence of speakers reanalysing a comitative phrase in a
monadic reciprocal construction as an argument. My analysis builds upon
recent work within LFG and is not only sensitive to these constructions'
diachronic development from a basic transitive verb, but also provides insight
into a variety of features these constructions share cross-linguistically.
Furthermore, this analysis predicts the syntactic behaviour of many naturally
symmetric verbs in English (such as dance, quarrel etc.).
2 Background
Many verbally marked reciprocal constructions have two forms; a basic
monadic form (1b) and a dyadic alternation (1c). The examples below are
from Swahili, but the monadic/dyadic alternation has been attested in many
other Bantu languages (e.g., Chicheŵa, Mchombo (1991) and Luganda,
McPherson (2008)) as well as Hungarian (Rákosi (2008)), Hebrew (Siloni
2001), Icelandic, German (Dimitriadis 2004) etc.:
(1a)
Swahili - Basic transitive construction2
Juma a-li-m-pig-a
Halima
Juma
3sgS-PST-3sgO-hit-FV Halima
“Juma hit Halima”
(1b)
Monadic reciprocal construction
Juma na Halima wa-li-pig-an-a
Juma and Halima 3plS-PST-hit-REC-FV
“Juma and Halima hit/fought each other”
(Both participants share roughly equal participation in this
event.)
1 Thank you to Mr Hillary Kissassi, my Swahili teacher, for his aid in the collection
of data presented here. For helpful comments and discussions, I would also like to
thank Rachel Nordlinger, Averil Grieve and György Rákosi. All errors are mine.
2 3sgS – 3rd person singular subject, REC – reciprocal morpheme, FV – final
vowel, MV – middle voice, APP – applicative.
291
(1c)
Dyadic reciprocal construction
Juma a-li-pig-an-a
na
Halima
Juma 3sgS-PST-hit-REC-FV
COM Halima
“Juma and Halima hit each other”
lit. “Juma hit.rec/fought with Halima”
(Unlike in (1b), Juma in this situation is usually the instigator
of the event and Halima's participation does not have to
match his – she may be a reluctant participant, for example.)
The monadic reciprocal construction (1b) is formed by marking the verb
with a reciprocal morpheme (-an-) and grouping the participants of the event
into the subject NP. The resulting reciprocated verb no longer has an object.
The dyadic construction (1c) is formed using the same reciprocal morphology
on the verb, but the subject is singular and in place of the object there is now
a prepositional phrase.
In previous work, Dimitriadis (2004) has convincingly argued for other
languages that the comitative phrase in these constructions is not part of a
discontinuous subject and that it is syntactically (and semantically) distinct
from the subject NP. In the case of Swahili, despite the fact that the
comitative preposition na shares the same form as the coordinator, the phrase
na Halima in (1c) can be shown to be distinct from the subject by verbal
agreement. Verbs in Swahili agree in number with their subject – irrespective
of whether the subject is discontinuous or not. So, unlike the dyadic
construction, the discontinuous form of the monadic construction has plural
subject agreement:
(2)
Discontinuous form of the monadic reciprocal construction
Juma wa-li-pig-an-a
na Halima
Juma 3plS-PST-hit-REC-FV and Halima
“Juma and Halima hit each other”
That (2) is distinct from the dyadic reciprocal construction in (1c) is also
demonstrated by the semantic differences between them. A dyadic reciprocal
construction allows the two participants to have unequal participation in the
event – however, (2) above has the same meaning as the monadic
construction in (1b) where both Halima and Juma are equal participants in the
event.
That the comitative phrase in (1c) is an argument (as opposed to an
adjunct) is demonstrated by the fact it is obligatory and that its existence is
presupposed by the verb:
292
(3)
*Juma a-li-pig-an-a
Juma 3sgS-PST-hit-REC-FV
*“Juma hit each other”
As far as I am aware, there is no account in LFG that unifies the
productive alternation between a transitive verb with its monadic and dyadic
reciprocal counterparts as occurs in Swahili. Previous analyses of the dyadic
construction have for the most part been limited to European languages
where the construction is restricted to a small set of verbs. Dimitriadis
(2004:4) for example, bases his analysis on the concept of irreducible
symmetry but he explicitly recognises that it cannot be applied to the Bantu
languages. Likewise, Rákosi (2008) provides an analysis that is
synchronically well-justified in the case of Hungarian monadic/dyadic
constructions – but it has limited cross-linguistic applicability because it
treats the verb in dyadic constructions as a basic lexeme that is no longer
derived from its transitive base.
What is needed for Swahili and the other Bantu languages is an account of
the monadic and dyadic reciprocal constructions that explains their
productive formation from a transitive base. I argue that the dyadic reciprocal
construction is formed by the reanalysis of a comitative phrase in a monadic
construction as a core argument of the verb. I capture this phenomenon in
LFG by building upon the work of two researchers; Rákosi (2008) and Webb
(2008). As a consequence of this analysis, not only do I provide an account of
the alternation of the Swahili monadic and dyadic constructions above with
respect to their transitive base, but in section 4 I also show how several
intriguing cross-linguistic properties of these constructions are predicted,
including:
1. Why the dyadic reciprocal construction is typically formed from a
comitative phrase.
2. Why these constructions have subtly different semantics with
respect to the participation of the entities in the symmetric relation.
3. Why the clitic marked reciprocal construction in Romance
languages, and periphrastic reciprocal construction in English (i.e.,
they saw each other) are not able to form the dyadic reciprocal
construction.
4. Why there exist in English a small set of so called “naturally
symmetric” verbs (such as dance, argue, fight etc. – see Kemmer
1993) which despite not being marked with reciprocal morphology
still allow the formation of a dyadic construction.
293
3 Analysis
In order to have cross-linguistic applicability, this analysis of the monadic
and dyadic reciprocal constructions in Swahili is built upon a new
understanding of comitative phrases. First though, I present an analysis of the
monadic construction as being formed from a transitive base. The dyadic
construction is in turn formed from the monadic reciprocal construction:
lexical base → monadic construction → dyadic construction
(1a) pig-a
(1b) pig-an-a
(1c) pig-an-a
na ...
hit
hit.rec
hit.rec / fight with
3.1
Formation of the monadic reciprocal construction
This analysis of the Swahili monadic construction ultimately draws from
work by Alsina (1996) in his account of reciprocal constructions in Catalan.
However, I differ from him by using the formalism introduced by Rákosi
(2008).3 The difference between these analyses is that Alsina maps two
argument slots to one grammatical function at the level of f-structure,
whereas Rákosi uses a process of “argument unification” to bundle two
arguments into a single slot at the level of a-structure, and this slot is
subsequently mapped to f-structure via the usual mapping principles. This
process occurs at the level of the lexicon – and can be understood as one that
takes a lexeme as its input, and produces another lexeme as its output. For
example, the a-structure of the verb piga – “hit” in (1a) above is:
(4)
piga < [P-A][P-P] >4
After undergoing argument unification, the two argument roles are
assigned a single argument slot. This produces the new lexeme in (1b) above,
pigana – “hit_each_other”, which like other lexemes forms a part of the
mental lexicon of the speaker:
(5)
pigana < [ [P-A][P-P] ] >
The outermost square brackets above indicate that the argument bundle of
proto-agent and proto-patient have been unified to a single argument slot.
For the mapping between a-structure and f-structure, I follow Rákosi in
3 Unlike Rákosi (2008), who forms the monadic reciprocal construction from the
dyadic construction, I will use this analysis to explain how the monadic
construction is formed from a transitive base.
4 I follow Alsina (1996) in using Dowty's (1993) analysis of thematic roles as
belonging to two prototypical categories: proto-agent (P-A) and proto-patient (PP).
294
employing Ackerman's (1992) version of lexical mapping theory. Ackerman
developed his version of mapping theory to incorporate Dowty's concept of
proto-roles.5 For reasons of space, I will only briefly outline how Ackerman's
theory differs from the LMT as described by Bresnan (2001):
Firstly, arguments are intrinsically classified by assigning them one of two
possible syntactic role features, [-r] or [-o]. The arguments are classified by
the following principles:
(1) The argument with the most heavily weighted proto-patient
properties is intrinsically classified as [-r].
(2) The argument with the most heavily weighted proto-agent properties
is intrinsically classified as [-o].
(3) All other arguments are intrinsically classified as [-o].
Ackerman (1992:64)
Secondly, arguments are assigned a default classification where the
highest ranking argument receives a [-r] feature and all other arguments
receive [+r], unless they have already received an intrinsic classification of
[-r]. The arguments are then mapped to grammatical functions according to
the standard principles of function-argument bi-uniqueness and the subject
condition (see Bresnan 2001:311 and the references contained within). For
example, consider the mapping between arguments and grammatical
functions for the transitive verb piga - “hit” and monadic reciprocal verb
pigana - “hit_each_other”:
Transitive hit from (1a)
Monadic hit.rec from (1b)
piga < [P-A][P-P] > →
pigana < [ [P-A][P-P] ] >
intr.
-o
-r
intr.
-o
def.
-r
def.
-r
bi-uniq.
SUBJ SUBJ/OBJ
bi-uniq.
SUBJ
result:
SUBJ OBJ
result:
SUBJ
The resulting analysis of the monadic reciprocal construction predicts that
it be intransitive, where the verb at the level of f-structure selects only a
subject. This prediction is borne out in (6) below where the clause is
ungrammatical because its associated f-structure violates the coherence
condition (see Bresnan 2001:63):
(6)
*Juma na Halima wa-li-pig-an-a
Juma and Halima 3plS-PST-hit-REC-FV
“Juma and Halima hit each other Fatuma”
Fatuma
Fatuma
5 Rákosi (2008) provides a very clear summary of the important features of
Ackerman's (1992) work, some of which are repeated here.
295
The analysis I present here for the formation of the monadic reciprocal
construction is similar to that of Rákosi's – the key difference being that it is
applied to a transitive verb rather than one in a dyadic reciprocal construction
(note in section 3.2.2 below I will return to Rákosi's work on Hungarian and
show how it can be encapsulated in this analysis of the dyadic reciprocal
construction). Insofar as this analysis presents the monadic reciprocal
construction as being formed via a lexical operation on a transitive base it is
akin to the work of Siloni (2008) and Reinhart and Siloni (2005).
3.2
Formation of the dyadic reciprocal construction
The most striking feature of dyadic reciprocal constructions from a crosslinguistic perspective is their formation from a comitative phrase. In fact, as
we will see in section 4, many of the unusual properties that dyadic reciprocal
constructions share in the world's languages actually arise as a consequence
of their relationship with comitative phrases. The analysis of the dyadic
reciprocal construction presented is able to explain these phenomena because
it builds upon a new understanding of comitative phrases in reciprocal
constructions which extends Webb's (2008) work on instrument phrases. In
section 3.2.1 below I first present my analysis of comitative phrases in LFG
before turning to the analysis of the Swahili dyadic reciprocal construction in
section 3.2.2.
3.2.1 Analysis of typical comitative phrases
A typical comitative phrase adds an optional entity to an event. This
comitative entity participates in the event in a manner very similar to another
participant in the event carrying a thematic role. For example, in (7) below,
the PP “with Abel” is a comitative phrase and Abel participates in the event
of catching in a manner similar to that of Cain:
(7)
Cain caught fish with Abel
The terminology I use here is that the comitative entity (Abel above) is
linked to another participant in the event. In this case Abel is linked to Cain
and his participation in the event is similar to that of Cain's. A further
examination of comitative phrases like those in (7) above reveal that they
typically have properties of both adjuncts and arguments:
Adjunct properties
• Comitative phrases are optional and their presence is not part of the
verb's meaning. In “Cain ate (spam) (with Abel)” the patient's
meaning is implied, even if not present, whereas the comitative's
meaning is wholly optional.
296
Argument properties
• Comitative phrases participate in the event described by the verb –
acting with another entity, and upon another entity, if present. For
example, in “Cain caught fish with Abel”, Abel has very similar
participation in this event as Cain.
• Comitative phrases are restricted in their number – unlike adjuncts.
So in (8) below, only one comitative phrase can appear after the verb
whereas in (9) many adjuncts may modify an event:
(8)
(9)
*Cain caught fish with Abel with Eve.
Cain caught fish in summer in the mornings.
As comitative phrases share properties of both adjuncts and arguments,
how should they be analysed? In fact, an analysis of phrases which sit in the
middle of the argument/adjunct continuum has been explored by many
linguists but most notably by Grimshaw (1990). She defines the term an “aadjunct” (argument/adjunct) as a phrase which has an intermediate status
between an argument and an adjunct and analyses it as:
(i)
(ii)
not being assigned a theta-role by the verb.
being licensed by the argument structure of the verb.
(adapted from Grimshaw 1990:108)
Her first property accounts for why the existence of comitatives is not
presupposed by the verb when they are not present (and hence their adjunctlike optionality). The second property accounts for why the number of
comitatives is restricted.
Webb (2008) analyses instruments as a-adjuncts and proposes that verbs
can have two tiers of argument structure; arguments on the first tier and aadjuncts on the second tier. I propose that comitative phrases are also usually
a-adjuncts and so should be analysed in a manner parallel to Webb's analysis
of instruments. For example, in (10) below, the 1 st tier argument structure has
two slots for the two thematic roles selected by the verb catch. However, an
optional 2nd tier of argument structure is also defined for the a-adjuncts:
(10)
Cain caught fish with Abel
catch: 1st tier < [P-A][P-P] >
2nd tier < [
] >
Note that in Webb's analysis, the instrument a-adjunct receives the thetarole of agent. I depart from his analysis here and follow Grimshaw (1990) in
supposing that comitative a-adjuncts are not directly assigned a theta-role.
Under the system of proto-roles adopted here (from Dowty 1993), this in turn
means that the comitative phrase is unspecified for a thematic role – instead
297
its participation in the event is understood to be the same as (or similar to)
that of another entity.6 I represent this below using empty square brackets [ ].
This analysis of a comitative entity as having an argument description of [ ]
ends up being equivalent to that which Rákosi gives for a partner thematic
role in his analysis of the dyadic reciprocal construction in Hungarian (see
Rákosi 2008:424 and the references contained within).
The mapping from argument structure to f-structure occurs via the
standard mapping rules described above, but with the provision that the
arguments in the first tier are mapped before any of the a-adjuncts in the
second tier:
catch: 1st tier < [P-A] [P-P] >
intrinsic
-o
-r
default
-r
bi-uniq.
SUBJ SUBJ/OBJ
result:
SUBJ OBJ
2nd tier < [ ] >
-o
+r
OBL
OBL
Below I outline some of the predictions this analysis makes which will
have relevance to my analysis of dyadic reciprocal constructions.
3.2.1.1 Comitative phrases as arguments
The first implication of this analysis is that we might expect to find verbs
which select a comitative phrase as an argument rather than as an a-adjunct,
especially given that a-adjuncts already appear at a lower tier of a-structure.
In this situation, the comitative entity is part of the verb's first tier of
argument structure – making it a prototypical argument. This is seen in
English with the verb quarrel:
(11)
Bob quarrelled with Fiona.
quarrel:
intrinsic
default
bi-uniq.
result:
1st tier < [P-A]
-o
-r
SUBJ
SUBJ
[ ] >
-o
+r
OBL
OBL
Note that this definition of quarrel is precisely that which Rákosi
(2008:444) gives for veszekedik - “quarrel” in Hungarian. This analysis also
raises the possibility of comitative phrases appearing as both a 1 st tier
argument and a 2nd tier a-adjunct. Rákosi (2003) observes that verbs which
6 Although a comitative entity is typically linked to the highest thematic role the
verb selects, this does not have to be the case. Consider “The man was beaten up
on the bus with his wife”. See Rákosi (2006:107-112) for further discussion.
298
treat the comitative phrase as being an argument optionally allow a
comitative a-adjunct as well – a possibility that is not available to verbs
which do not select a comitative phrase as an argument:7
(12a)
*Péter-rel (együtt) ritkán fut-ott-am Kati-val
Peter-with together rarely run-pst-1Sg Kate-with
*“I rarely ran with Kate with Peter”
(12b)
Péter-rel (együtt) ritkán veszeked-t-em Kati-val
Peter-with together rarely quarrel-pst-1Sg Kate-with
“I rarely quarrelled with Kate, together with Peter”
Rákosi (2003:3)
Under this analysis of comitative phrases, the verb veszekedtem –
“quarrel” in (12b) now selects a comitative entity in both 1 st and 2nd tier
argument structure:
veszekedtem
intrinsic
default
bi-uniq.
result:
1st tier < [P-A][ ] >
-o -o
-r +r
SUBJ OBL
SUBJ OBLarg
2nd tier < [
] >
-o
+r
OBL
OBLa-adj
3.2.1.2 Unequal participation in the event
In the analysis of comitative phrases above, we see that the comitative
entity does not receive the same thematic role as that of the entity to which it
is linked. This means we might expect that the comitative entity is able to
engage in the event in a manner different to that of its linked entity. This
behaviour is in fact observed in the day-to-day usage of comitative phrases.
For example, consider Maria's participation in the event in (13) below:
(13)
Maria selected photos for the gallery with Mark.
Although it is possible Maria and Mark have identical participation in the
event, my understanding is that Mark is more likely to be providing
assistance to Maria – perhaps looking over her shoulder. Fortunately, we do
not need to rely on intuitions alone to argue that there are differences between
these comitative entities and the entities to which they are linked – for a
restricted set of verbs, these intuitions can be formalised. For example,
consider the verb “collide”:8
7 An analogous example in English is “Michel fought [with Satan] Arg [with God on
his side]A-Adj”
8 Also dance – “Don danced with the broomstick”, fight – “The fishermen fought
with the elements” and others.
299
(14a)
(14b)
(14c)
The car collided with the tree.
#The car and the tree collided.
##The tree collided with the car.
It has been noted by many authors (e.g., Rákosi (2008), Dimitriadis
(2004), Gleitman et al. (1996), Dowty (1993) etc.) that sentences like (14b,c)
potentially throw doubt upon any analysis where the car and the tree must
participate identically in the predicate. Simply put, if both participants were
receiving identical thematic roles from the predicate, we would expect that
the participants could be swapped with little or no change in meaning – and
clearly that is not the case. Dimitriadis offers a counterargument to this point,
arguing that the differences in acceptability of (14b) and (14c) are caused by
“structural differences between the two argument positions” (Dimitriadis
2004:37). The evidence he uses to support this view comes from Gleitman et
al. (1996) who argue that even for predicates which must logically assign the
same thematic roles to two participants, there are still differences in
interpretation which seem dependent upon structural factors. For example
Gleitman et al. (1996) argue that both China and North Korea must have the
same thematic roles in (15) below – and so the differences in interpretation
between (15a) and (15b) are due to structural differences:
(15a)
(15b)
China is similar to North Korea.
North Korea is similar to China.
(from Dimitriadis 2004, based upon examples from Gleitman et al. 1996.)
Here, the ground (the oblique argument) is the source of comparison, and
the figure (the subject) is compared to it. For example, in (15b) North Korea
might be similar to China insofar as, like China, it has a communist
government. However, while structural effects might play some role in the
contrast in meaning between two structurally distinct arguments, the change
in acceptability of sentences like (14a) and (14b,c) above is factored around
agency – i.e., it is a thematic distinction. Consider again the contrast in
meaning between (14a) and (14c). Dowty (1993) points out that changes in
context change the admissibility of the sentences. For example, imagine the
following contexts: the tree has been torn from the ground by a hurricane and
slams into a stationary car, or the tree is animate (such as in a children's story)
and while running through the forest, trips over an unseen car. In these
contexts (14c) – “The tree collided with the car” is now the appropriate
description of the event. Examining the contexts in more detail we see that
the acceptability of (14c) is dependent upon whether the subject NP is
moving and/or animate. These are both properties that Dowty argues
contribute to the characterisation of an agent proto-role (Dowty 1993:572). In
other words, the tree and car in (14) have differing degrees of agency, and so
300
cannot carry identical thematic roles.
3.2.1.3 Concluding remarks on the comitative construction
This analysis of comitative phrases parallels that of Webb's analysis of
instruments. The typical comitative phrase is an a-adjunct which is
unspecified for a thematic role by the verb but whose presence is licensed in
a 2nd tier of argument structure. With respect to the analysis of the dyadic
reciprocal constructions, two properties of this analysis are of particular
importance:
1. That some predicates can select a comitative phrase as an argument
rather than an a-adjunct. This requires that the comitative entity be
selected in the first tier of argument structure rather than the second.
2. That the comitative entity need not have identical participation in the
event as that of the entity to which it is linked. This is as a result of
the comitative entity not being assigned a thematic role, rather its
participation in the event is inferred.
3.2.2 Analysis of the dyadic reciprocal construction in Swahili
I propose that in Swahili a comitative a-adjunct in the monadic reciprocal
construction has been reanalysed as an argument. Recall that Swahili
speakers productively form the monadic reciprocal construction from a
transitive base:
Transitive visit
tembelea < [P-A][P-P] >
→
Monadic visit.rec
tembeleana < [ [P-A][P-P] ] >
A verb in a monadic construction, like other verbs, can take a comitative
a-adjunct. For example, in (16) the comitative phrase na Daktari – “with the
doctor” is an a-adjunct and the a-structure of the verb tembeleana – “visit” is
given below:9
(16)
Juma na Halima wa-li-tembel-e-an-a
na
Daktari
Juma and Halima 3plS-PST-visit-APP-REC-FV COM doctor
“Juma and Halima visited each other, with the doctor”
Monadic visit.rec with a-adjunct
tembeleana:
1st tier < [ [P-A][P-P] ] >
2nd tier < [ ] >
Now consider the motivations for the reanalysis of the comitative a9 Unless given a specific context, this adjunct reading of (16) is now dispreferred in
favour of the argument reading of the comitative phrase where the doctor visited
Juma and Halima and they visited the doctor.
301
adjunct as an argument. The comitative entity adds a participant to the event
which engages in the same activity as another entity. Symmetric and lexically
derived reciprocal verbs have an interpretation that is very similar in its
entailments – two entities engage in an event with the same participation, but
because reciprocal verbs are symmetric, the verb now requires the presence
of a partner (e.g., for the verb hit in (1) above, both participants are hitting,
they are both being hit, and they are necessarily engaged in the event
together). Crucially then, those verbs which have an interpretation that is
symmetric at the level of the lexicon allow the possibility for a comitative aadjunct to be reanalysed as an argument in the 1st tier of argument structure.
Monadic visit.rec with a-adjunct
Dyadic visit.rec
tembeleana: 1st tier < [ [P-A][P-P] ] > → 1st tier < [ [P-A][P-P] ] [ ] >
2nd tier < [ ] >
Given that comitative phrases may already appear as 1 st tier arguments, it
is unsurprising that in the case of symmetric verbs a comitative a-adjunct can
be reanalysed in this way. Now that the comitative entity is an argument of
the verb, the subject NP no longer requires two participants as the symmetric
relation occurs between the subject NP and the comitative phrase:
(17)
Juma a-li-tembel-e-an-a
na
Halima
Juma 3sgS-PST-visit-APP-REC-FV COM Halima
“Juma and Halima visited each other”
This representation of reciprocal verbs in Swahili was built upon the
notion that speakers reanalyse a comitative a-adjunct in a monadic
construction as being an argument. The resulting form of the dyadic
reciprocated verb correctly predicts the syntax of these constructions. For
example, the a-structure of the dyadic verb pigana - “hit_each_other” from
(1c) maps the comitative entity to an oblique argument:
(18)
Dyadic pigana - “hit each other with” from (1c)
pigana: 1st tier < [ [P-A][P-P] ] [ ] >
intrinsic
-o
-o
default
-r
+r
bi-uniq.
SUBJ
OBL
result:
SUBJ
OBL
(1c)
Juma a-li-pig-an-a
na
Halima
Juma 3sgS-PST-hit-REC-FV
COM Halima
“Juma and Halima hit each other”
lit. “Juma hit.rec/fought with Halima”
302
Note that this final analysis of the Swahili dyadic construction is able to
provide a diachronic explanation for Rákosi's account of the same
construction in Hungarian. He argues that the dyadic reciprocal verb is basic,
or underived, with the monadic form of the verb being formed from the
dyadic base through the process of argument unification described above:
Rakosi's (2008) analysis for veszekedik - “quarrel”
basic dyadic lexeme
monadic form
veszekedik < [P-Agent] [
] >→
< [ [P-Agent] [
]] >
The most convincing evidence for his position is that the transitive
counterpart of many reciprocal verbs in Hungarian no longer exists or has a
meaning which is no longer related to the reciprocal form of the verb. The
limitation of this account is that it cannot (nor seeks to) explain the origin of
the dyadic verb and why it has a reciprocal morpheme. The analysis I present
here shows how this situation comes about. Recall that the reanalysis of a
comitative phrase as being an argument must occur at the level of the lexicon.
From a purely formal point of view, the dyadic reciprocation operation takes
a symmetric lexeme as its input and creates a new lexeme from it which
selects a comitative entity as an argument:
Verbrec,monadic 1st tier < [[P-A][P-P]] > → Verbrec,dyadic < [ [P-A][P-P] ] [ ] >
2nd tier < [ ] >
As these dyadic verbs are prototypically used with just two participants,
the complex structure of the first argument slot is a good candidate for
simplification during a process of lexicalisation:
Verbrec,dyadic < [ [P-A][P-P] ] [ ] >
→ Verbrec,dyadic,lexicalised < [P-A] [ ] >
This final version of the verb is now a bonafide new lexeme whose new astructure reflects that fact that speakers no longer consider it productively
formed from a transitive base.
3.3 Remarks
In this section I have argued that a comitative a-adjunct attached to a
reciprocated verb has been reanalysed as an argument. This reanalysis stems
from the fact that a reciprocated verb meets all the preconditions for
productive formation with a comitative entity, but with the additional
condition that the comitative entity no longer be optional. I have extended the
work of Rákosi (2008) and Webb (2008) to capture this phenomenon and I
have shown that this analysis correctly predicts the syntax of the monadic and
dyadic reciprocal constructions and the relationship they have with their
303
transitive base. However, the process I have described above is not specific to
just Swahili. In section 4 below I examine the implications this analysis has
from a cross-linguistic perspective and show that many properties of dyadic
reciprocal constructions can be now understood.
4 Cross-linguistic implications
4.1 Why the dyadic reciprocal construction is typically
formed from a comitative phrase
The analysis of dyadic reciprocal constructions presented above is built
upon the notion that a comitative a-adjunct is reanalysed as an argument.
Under this analysis, we would expect that a similar operation could occur in
different languages. More specifically though, it predicts that the dyadic
reciprocal construction be formed from a comitative phrase and
overwhelmingly, this is the case:
4.2
(19)
Greek (Hellenic) - Dimitriadis 2004:25
O Yanis filithike
me
ti
Maria
the John kissed-REC COM the Maria
lit. “John kissed each other with Maria”
“John and Maria kissed (each other)”
(20)
Hebrew (Afro-Asiatic) - Siloni 2008
Ha-yeladim hitnaešku
im
ha-yeladot
the-boys
kissed.REC COM the-girls
lit. “The boys kissed each other with the girls”
“The boys and girls kissed each other”
(21)
Hungarian (Uralic) - Rákosi 2008:443,444
A katoná-k
vesz-eked-t-ek
az ormester-rel
the soldier-pl quarrel-REC-PST-3pl the sergeant-COM
“The soldiers quarrelled with the sergeant”
(22)
Icelandic (Germanic)
Þeir
börðu-st
við
víkinga
3pl.m.nom fight.pl.PST-MV COM viking.m.pl.acc
“They fought (from berja - 'hit') with the vikings”
That the dyadic reciprocal construction is formed
from the monadic construction
Dimitriadis (2008, 2004) and Rákosi (2008) argue that the monadic
304
reciprocal construction is formed from its dyadic equivalent whereas the
analysis above predicts the opposite. Now, in the literature I have examined,
the monadic reciprocal constriction is either more prevalent than its dyadic
counterpart, or equally prevalent. Recall that both these constructions contain
reciprocal morphology, so any analysis which posits that the dyadic
construction is the base of formation for the monadic construction needs to
provide an additional mechanism to account for the existence of the monadic
constructions which have no dyadic counterpart. I think this is an unlikely
historical path for the formation of monadic reciprocal constructions.
4.3 Weakening of symmetric semantics
In this section I discuss the implications this analysis of dyadic reciprocal
constructions has on their associated semantics. In particular, I contrast the
predications it makes with Dimitriadis' (2004) work on irreducible symmetry
and show that some unexpected data revealed by Rákosi (2008) can be
explained by the analysis presented here.
4.3.1 Irreducible symmetry
Dimitriadis' notion of irreducible symmetry and its relation to dyadic
reciprocal constructions can be understood as the requirement that the
symmetric event be conceived as being atomic and that the symmetry be
absolute insofar as the entities are required to have identical participation
within the event. A predicate has the property of irreducible symmetry if:
(a)
(b)
it expresses a binary relationship, but
its two arguments have necessarily identical participation in
any event described by the predicate.
(Dimitriadis 2008:378)
For example, for the predicate kiss to be irreducibly symmetric requires
that both participants kiss is the same way in a single event. Situations where
one participant has a different participation in the event are ruled out, as are
situations which are only symmetric when seen as a collection of events (e.g.,
Henry kissed Sally on the hand at night, and Sally kissed Henry on the cheek
at dawn). Under the analysis presented here, the dyadic construction is
conceived of as arising from a reanalysis of a comitative a-adjunct as a
comitative argument. What semantic features might we expect in the
resulting construction – and are they compatible with Dimitriadis' conception
of irreducible symmetry? I characterised the comitative construction as
introducing an entity which participates in the same event as the entity to
which it is linked, but optionally with a less active participation. The
implications such an analysis has on our understanding on the formation of
dyadic reciprocal constructions revolve around two concepts; atomic events
305
and symmetry.
4.3.2 Atomic events
An atomic event is one where the participants engage in single event. For
example, in (23) below there is one event of cooking, and Olaf is doing it:
(23)
Olaf cooked some fish.
With a plural entity, we can often conceive of an event in two ways, as
being atomic or non-atomic. For example the event of cooking below can be
either atomic (24a) or non-atomic (24b):
(24a)
(24b)
Olaf and Sally cooked some fish together.
Olaf and Sally cooked some fish, Olaf in the morning and
Sally in the evening.
Comitative phrases introduce an entity to the event which is linked to
another participant. Above I argued that this linkage does not require that the
two entities have identical participation in the event, however, it does usually
require an atomic conception of the event. For example, like (24a) above, the
comitative construction in (25) requires an atomic interpretation of the event:
(25)
#Olaf cooked some fish with Sally, Olaf in the morning and
Sally in the evening.
With respect to reciprocity, unlike analytic (or phrasal) reciprocal
constructions, most monadic reciprocal constructions describe atomic events
because they are formed at the level of the lexicon. The addition of a
comitative entity further hardens this tendency because it is a participant
which shares in an atomic event. This view is in line with Dimitriadis – and is
supported by the massive amounts of linguistic evidence he provides.
4.3.3 Symmetric participation
Symmetric participation occurs when every participant in a relation acts as
both the initiator and endpoint of that relation. For example, we can say that
Olaf and Sally have symmetric participation in the event of seeing if we can
say: “Olaf saw Sally and Sally saw Olaf”. Symmetric participation is
independent of whether the event is atomic or not. For example, if Olaf saw
Sally in the morning, and Sally saw Olaf in the evening, their participation in
the relation of seeing is still symmetric. Given my analysis of comitative
entities, how might symmetry be affected in dyadic reciprocal constructions?
I argued above that comitative entities do not need to have identical
participation in the event with the entity to which they are linked – and often
have less agentive properties. Therefore we might expect dyadic
306
constructions to have weaker symmetric entailments between the two
participants. This predication is at odds with Dimitriadis (2008, 2004) who
argues that participation of the two entities in a dyadic construction must be
identical. However, my prediction is in line with the data where we do see
this weakening of symmetry – even to the point of allowing the symmetry to
be cancelled. For example, Rákosi (2008:423) illustrates this very clearly
with the following example:
(26)
Én num veszeked-t-em János-sal o veszeked-ett vel-em
I not quarrel-pst-1sg John-with he quarrel-pst with-1sg
“I was not quarrelling with John, he was quarrelling with me”
I have found similar examples for Icelandic (slást – “brawl” from the
middle voice of slá - “hit”) and this pattern is particularly clear in Swahili
where the symmetry can be cancelled for most verbs in the dyadic reciprocal
construction. For example:
(27)
Juma a-na-pend-an-a
na Halima,
Juma 3sgS-pres-love-rec-fv with Halima
lakini Halima ha-m-pend-i
Juma
but, Halima neg-3sgO-love-fv Juma
“Juma loves each other with Halima,
but Halima does not love Juma”
4.3.4 Concluding remarks
With respect to whether the symmetric situation should be seen as an
atomic event or as a collection of events, the analysis I present is in line with
work by Dimitriadis in predicting that dyadic reciprocal constructions be
atomic. However, contra Dimitriadis, I predict that the dyadic reciprocal
construction can have looser entailments with regard to symmetry when
compared to its monadic equivalent. This prediction is borne out in some
situations in Swahili and Hungarian where there are examples of the
comitative entity not participating in the symmetric event to the same extent
of the subject entity.
4.4
Lexical versus syntactic reciprocal constructions
Siloni (2008) notes that French, although having a superficial
resemblance to the Swahili and Hungarian reciprocal constructions, does not
allow the formation of a dyadic reciprocal construction:
(28)
*Jean s'-est embrassé avec Marie. (From Siloni 2008:482)
Jean se-is kissed
with Marie
*“Jean kissed each other with Marie”
307
Also, English:
(29)
*John saw each other with Marie.
The analysis I presented for the dyadic construction occurs at the level of
the lexicon where a comitative entity associated with a verb in a monadic
reciprocal construction is reanalysed as an argument. Because the reciprocal
construction in English is formed via a phrase, the symmetric sense of the
situation is introduced at the level of syntax – not at the level of the lexicon.
This means that the predicate in a reciprocal construction (see in (29) above)
is asymmetric at the level of the lexicon and consequently the comitative
phrase is not able to be reanalysed as an argument. Given that the dyadic
construction in French is ungrammatical, it would imply that the French
reciprocal construction is also formed at the level of syntax. Siloni (2008)
investigates these constructions in detail and this is her conclusion too. The
evidence she provides is based upon whether a reciprocal event can be
understood as a collection of asymmetric events or not. Her argument is that
lexemes can only denote atomic events, and the pluralisation of events must
occur at the level of syntax. Examples like (30) below demonstrate that the
French clitic reciprocal construction can describe a collection of asymmetric
events in the context of a game where Jean kisses Marie five times and Marie
kisses Jean five times (i.e., there were ten separate kisses):
(30)
Jean et Marie se sont embrassés cinq fois.
Jean and Marie se are kissed
five times
(From Siloni 2008:481)
Contrast this with the Swahili lexical reciprocal construction where the
only possible reading is that there were three shared kisses:
(31)
Juma na Halima wa-li-bus-i-an-a
mara tatu
Juma and Halima 3plS-PST-kiss-APP-REC-FV times three
“Juma and Halima kissed each other three times”
The analysis I present here fits with Siloni (2008) and Reinhart and
Siloni's (2005) work on reciprocals which treats the French reciprocal
construction as being syntactically derived, and which associates the dyadic
reciprocal construction with an operation that occurs at the level of the
lexicon.
4.5
Naturally Symmetric Verbs
In English there exist a set of “naturally symmetric” verbs (such as
quarrel, argue etc.) which appear to participate in a dyadic construction,
308
despite not having any reciprocal morphology. For example, consider (32)
below:
(32a)
(32b)
Bill and Mary argued/quarrelled/danced.
Bill argued/quarrelled/danced with Mary
This alternation is in fact predicated as a consequence of the formation of
the dyadic construction. Recall that this construction takes a symmetric verb
as its base and reanalyses an associated a-adjunct as an argument. Thus far
we have examined reciprocated (monadic) verbs which became symmetric as
a result of their formation. However, some verbs are already symmetric as
basic lexemes (such as quarrel etc.). These verbs then are suitable candidates
for participation in a dyadic construction because, like the derived symmetric
verbs in Swahili, they can serve as a base for the dyadic lexical operation.
5 Conclusion
The motivation for this analysis of the Swahili reciprocal constructions is
driven by a reanalysis of a comitative a-adjunct as an argument. Because
comitative constructions form part of the a-structure associated with a
predicate, this reanalysis necessarily occurs at the level of the lexicon.
Additionally, in order for the comitative a-adjunct to be reanalysed as an
argument, a candidate predicate requires the participation of two entities and
that these entities must engage in a relationship which is compatible with
comitative semantics. Both naturally symmetric verbs and symmetric verbs
that are formed in the lexicon fulfil these criteria; a symmetric verb requires
(prototypically) two participants with similar participation in the event (recall
for the Swahili verb pigana – “hit_each_other”, both participants are hitting,
they are both being hit, and they are engaged in the event together). This
understanding of comitative a-adjunct reanalysis predicts the central features
that lexically derived reciprocal constructions share cross-linguistically:
•
•
•
Why the entity represented by the oblique grammatical function is
typically marked as being a comitative phrase.
Why these constructions are so strongly associated with high degrees
of symmetry within a single event.
Why languages with reciprocal constructions formed in syntax are
not compatible with the dyadic reciprocal construction.
Finally, this explanation provides a natural account for why other
symmetric verbs which are not marked with reciprocal morphology (e.g.,
quarrel, bicker etc. in English or ongea - “chat” in Swahili) allow the same
alternations between a monadic and dyadic form.
309
6 References
Ackerman, F. 1992. Complex predicates and morphological relatedness:
locative alternation in Hungarian. In Sag, I. & Szabolcsi, A. (eds.),
Lexical Matters, 55-83. Stanford: CSLI Publications.
Alsina, A. 1996. The role of argument structure in grammar: Evidence from
Romance. Stanford: CSLI Publications.
Bresnan, J. 2001. Lexical-functional syntax. Blackwell Pub. Mass.
Dimitriadis, A. 2008. Irreducible symmetry in reciprocal constructions. In
König & Gast (eds.) 2008.
Dimitriadis, A. 2004. Discontinuous reciprocals. Manuscript: Utrecht
Institute of Linguistics OTS.
Dowty, D. 1993. Thematic proto-roles and argument selection. Language
67.3: 547-619.
Gleitman, L., Gleitman, H., Miller, C. & Ostrin, R. 1996. Similar, and similar
concepts. Cognition 58: 321-376.
Grimshaw, J. 1990. Argument structure. Cambridge, MA: The MIT Press.
Kemmer, S. 1993. The middle voice. John Benjamins.
Amsterdam/Philadelphia.
König, E. & Gast, V. (eds.) 2008. Reciprocals and reflexives: theoretical and
typological explorations. Mouton de Gruyter: Berlin, Germany.
Mchombo, S. 1991. Reciprocalization in Chichewa: A lexical account.
Linguistic Analysis 21:3-22 1991.
McPherson, L. 2008. A descriptive and theoretical account of Luganda
verbal morphology. BA thesis, Scripps College. Available online at
http://www.lcs.pomona.edu/McPherson_thesis.pdf.
Rákosi, G. 2008. The argument structure of inherently reflexive and
reciprocal predicates. In König & Gast 2008.
Rákosi, G. 2006. Dative experiencer predicates in Hungarian. PhD.
Dissertation. Utrecht. Published as volume 146 of the LOT Dissertation
Series. http://www.lotpublications.nl/publish/issues/Rakosi/index.html
Rákosi, G. 2003. Comitative arguments in Hungarian. In: Heeren, Willemijn,
Dimitra Papangeli and Evangelia Vlachou (eds.), UiL-OTS Yearbook
2003, 47-57. Utrecht: Utrecht Institute of Linguistics OTS.
Reinhart, T. & Siloni, T. 2005. The lexicon-syntax parameter. Linguistic
Inquiry Vol 36, Num 3, 2005.
Siloni, T. 2008. The syntax of reciprocal verbs: an overview. In König & Gast
2008.
Siloni, T. 2002. Active lexicon. Theoretical Linguistics, 28:383-400.
Siloni, T. 2001. Reciprocal verbs. In: Falk, Y. (ed.), Proceedings of the Israel
Association of Theoretical Linguistics.
Webb, J. 2008. Instruments in LFG's argument structure. Manuscript:
Master's thesis, University of Oxford. Available online at:
http://ucl.academia.edu/documents/0009/9899/Webb__MPhil_Thesis.pdf
310